A Guide to Neural Network Verification

Comprehensive guides covering foundational concepts to cutting-edge research techniques.

Phase 1

Foundations

Why verification matters, threat models, problem formulation, and theoretical fundamentals.

Learn NNV in 3 Minutes Why Neural Network Verification? The Verification Problem: Mathematical Formulation Threat Models in Neural Network Verification Soundness and Completeness Verification Taxonomy: A Systematic View Theoretical Barriers in Neural Network Verification

Phase 2

Methods & Tools

Core verification techniques including bound propagation, LP, SMT/MILP, and specialized methods.

Beyond ReLU: Modern Activation Functions Bound Propagation Approaches Branch-and-Bound Verification Linear Programming for Verification SMT and MILP Solvers for Verification Lipschitz Bounds and Curvature-Based Verification SDP-Based Verification: Semi-Definite Programming for Tighter Bounds Marabou and Reluplex: Extended Simplex for Verification Multi-Neuron Relaxation and PRIMA Neural Network Decomposition: The Unary-Binary Framework

Phase 3

Practice

Practical applications including robustness testing, certified training, benchmarks, and specifications.

Robustness Testing Guide Formal Specifications Falsifiers and Verifiers Verification Benchmarks Certified Adversarial Training Regularization-Based Robust Training

Phase 4

Advanced Topics

Cutting-edge research including scalability challenges, diverse architectures, and certified defenses.

Verification Scalability Verifying Diverse Architectures Training Robust Networks Certified Defenses and Randomized Smoothing Beyond Lp: Alternative Threat Models Real-World Applications